R&D Engineer at GRVC

Click on the title of each section to show/hide the content:

When

July 2008 to January 2013

Where

GRVC (Robotics Vision and Control group) Robotics research group of the University of Seville. web

Summary

Development of an inspection robot to search for cracks or leaks inside of the heat exchangers of a power plant: 15000 vertical pipes, 30 meters long, 200C, absolute darkness, humidity and a corrosive atmosphere... Thats a challenge!

I started this project when I was still a student. Part of the project was the Final Thesis of my Master Degree and later on I continued researching and developing a commercial version for two more years.

The project was developed for Iberdrola and it resulted in my first patent.

References

Rafael is one of the most creative persons, from the Engineering point of view, that I have known in my 16 years of experience as Professor. read letter

Jose Ramiro Martinez de Dios
Full professor at Universidad de Sevilla (My master thesis tutor)


I worked with Rafael for almost two years. I think that he is the most proactive and creative person I ever worked with. He always has a different and original idea for solving any issue. He loves finding new and better solutions to the problems. Additionally, he is a great event manager and he can handle big groups.

Arturo Torres Gonzalez
Robotics and Computer Vision Engineer (former colleague at GRVC)


Rafael Ferrin stands out for his brilliant ideas. Show him a problem and he will find ways to solve it that you would have never tought out.

Alberto de San Bernabe Clemente
Senior Manager Packaging Engineering APAC - Kenvue (former colleague at GRVC)


Rafael is a reliable, hard-worker person. He also has, not only analytical and problem solving skills, but also excellents interpersonal skills. He is always willing to learn new things so he is able to adapt to changes easily and quickly.

Luis Javier Duran LLanos
Senior Software Engineer II at Appian Corporation (former colleague at GRVC)


Even if our principal works were not related, Rafael was one of my contacts in his department. Every time that I had a problem that only his department could fix, Rafael has helped me in a fast, creative and efficient way. I saw him manage work under pressure and lead all kind of teams.
Thus, I want to recommend Rafael as researcher and wish him good luck in Japan.

Eduardo Ferrera
AI & Robotics Leader | Multi-Robot Systems & Industrial AI Expert (former colleague at GRVC)

Procedure

  • Analysis of the problem and current procedure
  • Analysis of the power plant working cycle and parameters (developed by another colleague)
  • Analysis of the limitations (space, temperature, humidity, corrosionE
  • Analysis of possible technologies (vibrations, conductivity, reflection, IR vision, visionE
  • Definition of the desired improvements and the technology selection criteria
  • Selection of the technology according to that criteria
  • Definition of the electromechanical system and components
  • Study of the geometry with a double intention:
    • available accessible space for the robot
    • Required observation of the environment and definition of the minimal necessary inspection paths
  • Definition of the mirrors and optical systems
  • Study of the illumination system
    • High illumination changes
    • Blinding reflections at mirrors and wet surfaces
    • Heat dissipation
  • Study of the vision system
    • Desired/required FOV (Field Of View)
    • Focus distances
    • Angle of observation
  • Simulation of the inspection system (27 hours of computer simulation and 4.2Gb of .txt files obtained)
  • Data mining of the obtained information
    • Ratio of welding lines observed zero, once, twice or several times for each configuration and defined paths
    • Detection of critical situations and black areas of the map.
    • Definition of the number of required cameras and optimal observing angles.
    • Observation distances
    • Selection of inspection priorities; simplifications.
    • Estimation of inspection times.
  • Creation of the first prototype
    • Environment mock up.
    • Mechanical design.
    • Optical design.
    • Electronic design (developed by another colleague)
    • Communications design (developed by another colleague)
    • Software design (developed by another colleague)
    • Refrigeration design (developed by another colleague)
  • Analysis of the captures of the prototype
    • Image deformation due to the mirrors
    • Optical system simulation
    • Image transformation matrix for image correction
  • Computer vision
    • Standard filters for computer vision
    • Custom border detection filters
    • Hough transformation
    • Custom Hough-based transformation for projecting lines
  • Design of the second prototype
    • Mechanical redesign, including new functionalities and production oriented design.
    • Optical Improvement.
    • Electronic re-design (using Maxon Motors commercial controllers)
    • Communications re-design (developed by another colleague)
    • Software re-design (developed by another colleague)
    • Refrigeration improvement (developed by another colleague)

Skills

  • Matlab
    I used Matlab for creating a geometrical simulation of the environment and analyse the visible parts of the pipes from different paths. Later on, with the results of that simulation, I extracted extra useful information like the usual observation distances for different parts of the pipes, the optimal inspection angles and FOVs, etc. I also used Matlab to create a simulation of the optical system (with plain and spherical mirrors) to determine the original position on the image plane of each pixel of the camera, and later on I created a transformation map to correct that deformation.
  • Computer Vision
    Also in Matlab, I applied basic image filters to correct contrast and clean some noises to prepare the image for border detection. I tried different standard filters for border detection and later on I created a custom one for our application, where the borders to detect are always lines with a common point (the vanishing point of the vertical lines). The same about Hough transformation; instead of two parameters for each line (ax+b), I created a custom version to detect lines from a same family (vertical lines in perspective: the same vanishing point). The good point was that instead of changing the geometry of the line each time, I created a set of filters for lines with the same origin, and then my parameters was the position of the origin. Changing the origin was as simple as creating an offset when applying the filter to the image. Later I realise that it was more accurate integrate both border detection and line detections in the same filter, reducing computational time and optimising the border detection to exactly the line we was looking for.
  • CATIA
    The first Mechanical design was thought by me, but built into Solid Edge by a colleague. Later I started learning CATIA and the second prototype (15 months later) was developed by me on CATIA. Still the design was not enough detailed for manufacturing, so the company we hired for that improved my design to make it manufacturable. Later on I have improve a lot my knowledge about CATIA. For more details, read my experience at FACC.